Local-as-view mediators allow to query semantic heterogeneous data sources such as linked data endpoints, open data servers or any available web service. Nevertheless, the problem of rewriting conjunctive queries on LAV mediators is intractable in general, and the number of rewritings can be exponential on the size of the query. Most important, in the context of the Semantic Web even simple conjunctive SPARQL queries can be rewritten in millions of rewritings. In this paper, we propose SemLAV, a LAV mediation technique that relies on Semantic Web technologies to execute SPARQL queries against data sources. SemLAV implements a polynomial view selection algorithm to choose the relevant views that can answer a given SPARQL query. Additionally, SemLAV implements an execution strategy for gathering data from the relevant views and outputs a great portion of the answer in a short time. We present an experimental study for SemLAV, and compare its performance with execution strategies for LAV rewritings. The results suggest that SemLAV scales up to SPARQL queries and a large number of views, while it significantly outperforms existing solutions.